In [1]:
import matplotlib.pyplot as plt
import numpy as np
import PIL
import tensorflow as tf

from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential
In [2]:
import pathlib
data_dir = 'C:\\Users\\starl\\Downloads\\class\\data2\\'
data_dir = pathlib.Path(data_dir).with_suffix('')
In [3]:
data_dir
Out[3]:
WindowsPath('C:/Users/starl/Downloads/class/data2')
In [4]:
image_count = len(list(data_dir.glob('*/*.png')))
print(image_count)
# 데이터 사본의 총 개수

234
In [8]:
ventis = list(data_dir.glob('Venti/*'))
PIL.Image.open(str(ventis[0]))
Out[8]:
No description has been provided for this image
In [10]:
kokomi = list(data_dir.glob('Sangonomiya Kokomi/*'))
PIL.Image.open(str(kokomi[0]))
Out[10]:
No description has been provided for this image
In [11]:
# 데이터셋 생성 ( 매개변수 지정 )
batch_size = 32
img_height = 180
img_width = 180
In [12]:
train_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="training",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

Found 234 files belonging to 60 classes.
Using 188 files for training.
In [13]:
val_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="validation",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

Found 234 files belonging to 60 classes.
Using 46 files for validation.
In [14]:
class_names = train_ds.class_names
print(class_names)

['Aether', 'Albedo', 'Aloy', 'Amber', 'Arataki Itto', 'Barbara', 'Beidou', 'Bennett', 'Candace', 'Chongyun', 'Collei', 'Cyno', 'Diluc', 'Diona', 'Dori', 'Eula', 'Fischl', 'Ganyu', 'Gorou', 'Hu Tao', 'Jean', 'Kaedehara Kazuha', 'Kaeya', 'Kamisato Ayaka', 'Kamisato Ayato', 'Keqing', 'Klee', 'Kujou Sara', 'Kuki Shinobu', 'Layla', 'Lisa', 'Lumine', 'Mona', 'Nahida', 'Nilou', 'Ningguang', 'Noelle', 'Qiqi', 'Raiden Shogun', 'Razor', 'Rosaria', 'Sangonomiya Kokomi', 'Sayu', 'Shenhe', 'Shikanoin Heizou', 'Sucrose', 'Tartaglia', 'Thoma', 'Tighnari', 'Venti', 'Xiangling', 'Xiao', 'Xingqiu', 'Xinyan', 'Yae Miko', 'Yanfei', 'Yelan', 'Yoimiya', 'Yun Jin', 'Zhongli']
In [15]:
import matplotlib.pyplot as plt

plt.figure(figsize=(10, 10))
for images, labels in train_ds.take(1):
  for i in range(9):
    ax = plt.subplot(5, 10, i + 1)
    plt.imshow(images[i].numpy().astype("uint8"))
    plt.title(class_names[labels[i]])
    plt.axis("off")
    # 훈련 데이터세트 출력
No description has been provided for this image
In [16]:
for image_batch, labels_batch in train_ds:
  print(image_batch.shape)
  print(labels_batch.shape)
  break
# image_batch는 (32, 180, 180, 3) 형상의 텐서

(32, 180, 180, 3)
(32,)
In [17]:
# 데이터세트 구성
AUTOTUNE = tf.data.AUTOTUNE

train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
In [18]:
# tf.keras.layers.Rescaling을 사용하여 값을 [0, 1] 범위로 표준화
normalization_layer = layers.Rescaling(1./255)
In [19]:
normalized_ds = train_ds.map(lambda x, y: (normalization_layer(x), y))
image_batch, labels_batch = next(iter(normalized_ds))
first_image = image_batch[0]
print(np.min(first_image), np.max(first_image))
# Dataset.map을 호출하여 데이터세트에 이를 적용

0.0 1.0
In [20]:
# 모델 생성
num_classes = len(class_names)

model = Sequential([
  layers.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(64, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Flatten(),
  layers.Dense(128, activation='relu'),
  layers.Dense(num_classes)
])
# relu 활성화 함수를 통해서 레이어 생성
In [21]:
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
# 아담 옵티마이저와 tf.keras.losses.SparseCategoricalCrossentropy 손실 함수를 선택
In [22]:
model.summary()
# 네트워크의 모든 레이어 호출( df['total_bill'].info() 같은 메서드 )

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 rescaling_1 (Rescaling)     (None, 180, 180, 3)       0         
                                                                 
 conv2d (Conv2D)             (None, 180, 180, 16)      448       
                                                                 
 max_pooling2d (MaxPooling2  (None, 90, 90, 16)        0         
 D)                                                              
                                                                 
 conv2d_1 (Conv2D)           (None, 90, 90, 32)        4640      
                                                                 
 max_pooling2d_1 (MaxPoolin  (None, 45, 45, 32)        0         
 g2D)                                                            
                                                                 
 conv2d_2 (Conv2D)           (None, 45, 45, 64)        18496     
                                                                 
 max_pooling2d_2 (MaxPoolin  (None, 22, 22, 64)        0         
 g2D)                                                            
                                                                 
 flatten (Flatten)           (None, 30976)             0         
                                                                 
 dense (Dense)               (None, 128)               3965056   
                                                                 
 dense_1 (Dense)             (None, 60)                7740      
                                                                 
=================================================================
Total params: 3996380 (15.24 MB)
Trainable params: 3996380 (15.24 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
In [23]:
# 100에포크 동안 모델 학습
epochs=100
history = model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs
)

Epoch 1/100
6/6 [==============================] - 9s 1s/step - loss: 4.3825 - accuracy: 0.0106 - val_loss: 4.1407 - val_accuracy: 0.0000e+00
Epoch 2/100
6/6 [==============================] - 6s 1s/step - loss: 4.0310 - accuracy: 0.0638 - val_loss: 4.1804 - val_accuracy: 0.0217
Epoch 3/100
6/6 [==============================] - 5s 788ms/step - loss: 3.8202 - accuracy: 0.1702 - val_loss: 4.2850 - val_accuracy: 0.0000e+00
Epoch 4/100
6/6 [==============================] - 5s 861ms/step - loss: 3.4340 - accuracy: 0.2287 - val_loss: 4.4371 - val_accuracy: 0.0217
Epoch 5/100
6/6 [==============================] - 7s 1s/step - loss: 2.7521 - accuracy: 0.4628 - val_loss: 4.9281 - val_accuracy: 0.0000e+00
Epoch 6/100
6/6 [==============================] - 7s 1s/step - loss: 1.6472 - accuracy: 0.6543 - val_loss: 5.5912 - val_accuracy: 0.0217
Epoch 7/100
6/6 [==============================] - 5s 946ms/step - loss: 0.8228 - accuracy: 0.8723 - val_loss: 6.2738 - val_accuracy: 0.0217
Epoch 8/100
6/6 [==============================] - 6s 1s/step - loss: 0.4674 - accuracy: 0.9043 - val_loss: 7.1846 - val_accuracy: 0.0435
Epoch 9/100
6/6 [==============================] - 7s 1s/step - loss: 0.1745 - accuracy: 0.9787 - val_loss: 8.2793 - val_accuracy: 0.0217
Epoch 10/100
6/6 [==============================] - 7s 1s/step - loss: 0.0827 - accuracy: 0.9840 - val_loss: 8.7844 - val_accuracy: 0.0217
Epoch 11/100
6/6 [==============================] - 6s 991ms/step - loss: 0.0250 - accuracy: 0.9947 - val_loss: 9.3569 - val_accuracy: 0.0217
Epoch 12/100
6/6 [==============================] - 6s 1s/step - loss: 0.0029 - accuracy: 1.0000 - val_loss: 9.7865 - val_accuracy: 0.0652
Epoch 13/100
6/6 [==============================] - 4s 650ms/step - loss: 0.0046 - accuracy: 1.0000 - val_loss: 10.1966 - val_accuracy: 0.0652
Epoch 14/100
6/6 [==============================] - 4s 645ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 10.1790 - val_accuracy: 0.0652
Epoch 15/100
6/6 [==============================] - 5s 800ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 10.1164 - val_accuracy: 0.0652
Epoch 16/100
6/6 [==============================] - 4s 700ms/step - loss: 8.3651e-04 - accuracy: 1.0000 - val_loss: 10.1437 - val_accuracy: 0.0652
Epoch 17/100
6/6 [==============================] - 5s 894ms/step - loss: 5.4617e-04 - accuracy: 1.0000 - val_loss: 10.1915 - val_accuracy: 0.0652
Epoch 18/100
6/6 [==============================] - 4s 736ms/step - loss: 3.7098e-04 - accuracy: 1.0000 - val_loss: 10.3012 - val_accuracy: 0.0652
Epoch 19/100
6/6 [==============================] - 5s 940ms/step - loss: 2.7360e-04 - accuracy: 1.0000 - val_loss: 10.3904 - val_accuracy: 0.0652
Epoch 20/100
6/6 [==============================] - 5s 872ms/step - loss: 2.1921e-04 - accuracy: 1.0000 - val_loss: 10.4642 - val_accuracy: 0.0652
Epoch 21/100
6/6 [==============================] - 5s 736ms/step - loss: 1.8513e-04 - accuracy: 1.0000 - val_loss: 10.5173 - val_accuracy: 0.0652
Epoch 22/100
6/6 [==============================] - 5s 828ms/step - loss: 1.6289e-04 - accuracy: 1.0000 - val_loss: 10.5631 - val_accuracy: 0.0652
Epoch 23/100
6/6 [==============================] - 6s 1s/step - loss: 1.4345e-04 - accuracy: 1.0000 - val_loss: 10.5925 - val_accuracy: 0.0652
Epoch 24/100
6/6 [==============================] - 6s 967ms/step - loss: 1.2964e-04 - accuracy: 1.0000 - val_loss: 10.6117 - val_accuracy: 0.0652
Epoch 25/100
6/6 [==============================] - 4s 765ms/step - loss: 1.1864e-04 - accuracy: 1.0000 - val_loss: 10.6293 - val_accuracy: 0.0652
Epoch 26/100
6/6 [==============================] - 5s 901ms/step - loss: 1.0985e-04 - accuracy: 1.0000 - val_loss: 10.6466 - val_accuracy: 0.0652
Epoch 27/100
6/6 [==============================] - 5s 869ms/step - loss: 1.0258e-04 - accuracy: 1.0000 - val_loss: 10.6617 - val_accuracy: 0.0652
Epoch 28/100
6/6 [==============================] - 4s 713ms/step - loss: 9.7138e-05 - accuracy: 1.0000 - val_loss: 10.6804 - val_accuracy: 0.0652
Epoch 29/100
6/6 [==============================] - 5s 826ms/step - loss: 8.9901e-05 - accuracy: 1.0000 - val_loss: 10.6948 - val_accuracy: 0.0652
Epoch 30/100
6/6 [==============================] - 7s 1s/step - loss: 8.5391e-05 - accuracy: 1.0000 - val_loss: 10.7104 - val_accuracy: 0.0652
Epoch 31/100
6/6 [==============================] - 5s 775ms/step - loss: 8.0914e-05 - accuracy: 1.0000 - val_loss: 10.7297 - val_accuracy: 0.0652
Epoch 32/100
6/6 [==============================] - 6s 962ms/step - loss: 7.6429e-05 - accuracy: 1.0000 - val_loss: 10.7433 - val_accuracy: 0.0652
Epoch 33/100
6/6 [==============================] - 4s 710ms/step - loss: 7.2769e-05 - accuracy: 1.0000 - val_loss: 10.7580 - val_accuracy: 0.0652
Epoch 34/100
6/6 [==============================] - 4s 743ms/step - loss: 6.9906e-05 - accuracy: 1.0000 - val_loss: 10.7763 - val_accuracy: 0.0652
Epoch 35/100
6/6 [==============================] - 4s 747ms/step - loss: 6.6448e-05 - accuracy: 1.0000 - val_loss: 10.7906 - val_accuracy: 0.0652
Epoch 36/100
6/6 [==============================] - 5s 852ms/step - loss: 6.3763e-05 - accuracy: 1.0000 - val_loss: 10.8084 - val_accuracy: 0.0652
Epoch 37/100
6/6 [==============================] - 5s 811ms/step - loss: 6.0624e-05 - accuracy: 1.0000 - val_loss: 10.8219 - val_accuracy: 0.0652
Epoch 38/100
6/6 [==============================] - 6s 1s/step - loss: 5.8421e-05 - accuracy: 1.0000 - val_loss: 10.8361 - val_accuracy: 0.0652
Epoch 39/100
6/6 [==============================] - 4s 721ms/step - loss: 5.6304e-05 - accuracy: 1.0000 - val_loss: 10.8518 - val_accuracy: 0.0652
Epoch 40/100
6/6 [==============================] - 5s 882ms/step - loss: 5.3717e-05 - accuracy: 1.0000 - val_loss: 10.8638 - val_accuracy: 0.0652
Epoch 41/100
6/6 [==============================] - 5s 798ms/step - loss: 5.1974e-05 - accuracy: 1.0000 - val_loss: 10.8792 - val_accuracy: 0.0652
Epoch 42/100
6/6 [==============================] - 6s 1s/step - loss: 4.9936e-05 - accuracy: 1.0000 - val_loss: 10.8925 - val_accuracy: 0.0652
Epoch 43/100
6/6 [==============================] - 5s 848ms/step - loss: 4.8110e-05 - accuracy: 1.0000 - val_loss: 10.9035 - val_accuracy: 0.0652
Epoch 44/100
6/6 [==============================] - 5s 871ms/step - loss: 4.6562e-05 - accuracy: 1.0000 - val_loss: 10.9157 - val_accuracy: 0.0652
Epoch 45/100
6/6 [==============================] - 7s 1s/step - loss: 4.4760e-05 - accuracy: 1.0000 - val_loss: 10.9268 - val_accuracy: 0.0652
Epoch 46/100
6/6 [==============================] - 6s 1s/step - loss: 4.3376e-05 - accuracy: 1.0000 - val_loss: 10.9359 - val_accuracy: 0.0652
Epoch 47/100
6/6 [==============================] - 5s 786ms/step - loss: 4.1922e-05 - accuracy: 1.0000 - val_loss: 10.9461 - val_accuracy: 0.0652
Epoch 48/100
6/6 [==============================] - 5s 770ms/step - loss: 4.0665e-05 - accuracy: 1.0000 - val_loss: 10.9568 - val_accuracy: 0.0652
Epoch 49/100
6/6 [==============================] - 5s 970ms/step - loss: 3.9349e-05 - accuracy: 1.0000 - val_loss: 10.9674 - val_accuracy: 0.0652
Epoch 50/100
6/6 [==============================] - 4s 717ms/step - loss: 3.8422e-05 - accuracy: 1.0000 - val_loss: 10.9806 - val_accuracy: 0.0652
Epoch 51/100
6/6 [==============================] - 5s 842ms/step - loss: 3.7213e-05 - accuracy: 1.0000 - val_loss: 10.9925 - val_accuracy: 0.0652
Epoch 52/100
6/6 [==============================] - 5s 830ms/step - loss: 3.6017e-05 - accuracy: 1.0000 - val_loss: 11.0017 - val_accuracy: 0.0652
Epoch 53/100
6/6 [==============================] - 5s 798ms/step - loss: 3.5097e-05 - accuracy: 1.0000 - val_loss: 11.0107 - val_accuracy: 0.0652
Epoch 54/100
6/6 [==============================] - 5s 791ms/step - loss: 3.4071e-05 - accuracy: 1.0000 - val_loss: 11.0211 - val_accuracy: 0.0652
Epoch 55/100
6/6 [==============================] - 5s 789ms/step - loss: 3.3137e-05 - accuracy: 1.0000 - val_loss: 11.0312 - val_accuracy: 0.0652
Epoch 56/100
6/6 [==============================] - 5s 781ms/step - loss: 3.2218e-05 - accuracy: 1.0000 - val_loss: 11.0378 - val_accuracy: 0.0652
Epoch 57/100
6/6 [==============================] - 5s 829ms/step - loss: 3.1417e-05 - accuracy: 1.0000 - val_loss: 11.0470 - val_accuracy: 0.0652
Epoch 58/100
6/6 [==============================] - 7s 1s/step - loss: 3.0649e-05 - accuracy: 1.0000 - val_loss: 11.0573 - val_accuracy: 0.0652
Epoch 59/100
6/6 [==============================] - 6s 942ms/step - loss: 2.9817e-05 - accuracy: 1.0000 - val_loss: 11.0654 - val_accuracy: 0.0652
Epoch 60/100
6/6 [==============================] - 5s 787ms/step - loss: 2.9164e-05 - accuracy: 1.0000 - val_loss: 11.0756 - val_accuracy: 0.0652
Epoch 61/100
6/6 [==============================] - 5s 919ms/step - loss: 2.8411e-05 - accuracy: 1.0000 - val_loss: 11.0855 - val_accuracy: 0.0652
Epoch 62/100
6/6 [==============================] - 6s 931ms/step - loss: 2.7772e-05 - accuracy: 1.0000 - val_loss: 11.0938 - val_accuracy: 0.0652
Epoch 63/100
6/6 [==============================] - 5s 790ms/step - loss: 2.7037e-05 - accuracy: 1.0000 - val_loss: 11.1018 - val_accuracy: 0.0652
Epoch 64/100
6/6 [==============================] - 4s 727ms/step - loss: 2.6436e-05 - accuracy: 1.0000 - val_loss: 11.1087 - val_accuracy: 0.0652
Epoch 65/100
6/6 [==============================] - 4s 704ms/step - loss: 2.5894e-05 - accuracy: 1.0000 - val_loss: 11.1167 - val_accuracy: 0.0652
Epoch 66/100
6/6 [==============================] - 5s 780ms/step - loss: 2.5258e-05 - accuracy: 1.0000 - val_loss: 11.1239 - val_accuracy: 0.0652
Epoch 67/100
6/6 [==============================] - 5s 753ms/step - loss: 2.4751e-05 - accuracy: 1.0000 - val_loss: 11.1326 - val_accuracy: 0.0652
Epoch 68/100
6/6 [==============================] - 5s 773ms/step - loss: 2.4191e-05 - accuracy: 1.0000 - val_loss: 11.1411 - val_accuracy: 0.0652
Epoch 69/100
6/6 [==============================] - 4s 739ms/step - loss: 2.3672e-05 - accuracy: 1.0000 - val_loss: 11.1495 - val_accuracy: 0.0652
Epoch 70/100
6/6 [==============================] - 4s 725ms/step - loss: 2.3168e-05 - accuracy: 1.0000 - val_loss: 11.1561 - val_accuracy: 0.0652
Epoch 71/100
6/6 [==============================] - 5s 791ms/step - loss: 2.2704e-05 - accuracy: 1.0000 - val_loss: 11.1644 - val_accuracy: 0.0652
Epoch 72/100
6/6 [==============================] - 4s 663ms/step - loss: 2.2221e-05 - accuracy: 1.0000 - val_loss: 11.1712 - val_accuracy: 0.0652
Epoch 73/100
6/6 [==============================] - 4s 648ms/step - loss: 2.1765e-05 - accuracy: 1.0000 - val_loss: 11.1784 - val_accuracy: 0.0652
Epoch 74/100
6/6 [==============================] - 4s 664ms/step - loss: 2.1369e-05 - accuracy: 1.0000 - val_loss: 11.1870 - val_accuracy: 0.0652
Epoch 75/100
6/6 [==============================] - 4s 686ms/step - loss: 2.0982e-05 - accuracy: 1.0000 - val_loss: 11.1962 - val_accuracy: 0.0652
Epoch 76/100
6/6 [==============================] - 4s 707ms/step - loss: 2.0544e-05 - accuracy: 1.0000 - val_loss: 11.2038 - val_accuracy: 0.0652
Epoch 77/100
6/6 [==============================] - 4s 725ms/step - loss: 2.0097e-05 - accuracy: 1.0000 - val_loss: 11.2096 - val_accuracy: 0.0652
Epoch 78/100
6/6 [==============================] - 5s 820ms/step - loss: 1.9732e-05 - accuracy: 1.0000 - val_loss: 11.2160 - val_accuracy: 0.0652
Epoch 79/100
6/6 [==============================] - 5s 923ms/step - loss: 1.9368e-05 - accuracy: 1.0000 - val_loss: 11.2237 - val_accuracy: 0.0652
Epoch 80/100
6/6 [==============================] - 5s 799ms/step - loss: 1.9013e-05 - accuracy: 1.0000 - val_loss: 11.2306 - val_accuracy: 0.0652
Epoch 81/100
6/6 [==============================] - 5s 759ms/step - loss: 1.8672e-05 - accuracy: 1.0000 - val_loss: 11.2388 - val_accuracy: 0.0652
Epoch 82/100
6/6 [==============================] - 5s 818ms/step - loss: 1.8323e-05 - accuracy: 1.0000 - val_loss: 11.2461 - val_accuracy: 0.0652
Epoch 83/100
6/6 [==============================] - 4s 740ms/step - loss: 1.7994e-05 - accuracy: 1.0000 - val_loss: 11.2529 - val_accuracy: 0.0652
Epoch 84/100
6/6 [==============================] - 4s 690ms/step - loss: 1.7682e-05 - accuracy: 1.0000 - val_loss: 11.2590 - val_accuracy: 0.0652
Epoch 85/100
6/6 [==============================] - 4s 666ms/step - loss: 1.7376e-05 - accuracy: 1.0000 - val_loss: 11.2665 - val_accuracy: 0.0652
Epoch 86/100
6/6 [==============================] - 4s 725ms/step - loss: 1.7066e-05 - accuracy: 1.0000 - val_loss: 11.2730 - val_accuracy: 0.0652
Epoch 87/100
6/6 [==============================] - 4s 690ms/step - loss: 1.6743e-05 - accuracy: 1.0000 - val_loss: 11.2780 - val_accuracy: 0.0652
Epoch 88/100
6/6 [==============================] - 4s 735ms/step - loss: 1.6459e-05 - accuracy: 1.0000 - val_loss: 11.2847 - val_accuracy: 0.0652
Epoch 89/100
6/6 [==============================] - 4s 696ms/step - loss: 1.6186e-05 - accuracy: 1.0000 - val_loss: 11.2908 - val_accuracy: 0.0652
Epoch 90/100
6/6 [==============================] - 5s 792ms/step - loss: 1.5968e-05 - accuracy: 1.0000 - val_loss: 11.2985 - val_accuracy: 0.0652
Epoch 91/100
6/6 [==============================] - 4s 760ms/step - loss: 1.5643e-05 - accuracy: 1.0000 - val_loss: 11.3056 - val_accuracy: 0.0652
Epoch 92/100
6/6 [==============================] - 4s 756ms/step - loss: 1.5380e-05 - accuracy: 1.0000 - val_loss: 11.3112 - val_accuracy: 0.0652
Epoch 93/100
6/6 [==============================] - 4s 727ms/step - loss: 1.5141e-05 - accuracy: 1.0000 - val_loss: 11.3178 - val_accuracy: 0.0652
Epoch 94/100
6/6 [==============================] - 4s 689ms/step - loss: 1.4902e-05 - accuracy: 1.0000 - val_loss: 11.3247 - val_accuracy: 0.0652
Epoch 95/100
6/6 [==============================] - 4s 707ms/step - loss: 1.4646e-05 - accuracy: 1.0000 - val_loss: 11.3319 - val_accuracy: 0.0652
Epoch 96/100
6/6 [==============================] - 5s 798ms/step - loss: 1.4395e-05 - accuracy: 1.0000 - val_loss: 11.3380 - val_accuracy: 0.0652
Epoch 97/100
6/6 [==============================] - 4s 734ms/step - loss: 1.4159e-05 - accuracy: 1.0000 - val_loss: 11.3430 - val_accuracy: 0.0652
Epoch 98/100
6/6 [==============================] - 5s 781ms/step - loss: 1.3937e-05 - accuracy: 1.0000 - val_loss: 11.3476 - val_accuracy: 0.0652
Epoch 99/100
6/6 [==============================] - 5s 846ms/step - loss: 1.3708e-05 - accuracy: 1.0000 - val_loss: 11.3521 - val_accuracy: 0.0652
Epoch 100/100
6/6 [==============================] - 6s 994ms/step - loss: 1.3516e-05 - accuracy: 1.0000 - val_loss: 11.3587 - val_accuracy: 0.0652
In [25]:
model.save('./models/genshin.HDF5')

INFO:tensorflow:Assets written to: ./models/genshin.HDF5\assets

INFO:tensorflow:Assets written to: ./models/genshin.HDF5\assets
In [26]:
# 훈련 결과 시각화
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']

loss = history.history['loss']
val_loss = history.history['val_loss']

epochs_range = range(epochs)

plt.figure(figsize=(8, 8))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
No description has been provided for this image
In [24]:
PIL.Image.open('./test/venti.jpg')
Out[24]:
No description has been provided for this image
In [31]:
# 모델 테스트
char_path = './test/venti.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 109ms/step
이 캐릭터는 [ Eula ] 같습니다. 추정 확률은 26.56 % 입니다.
In [28]:
PIL.Image.open('./test/kokomi.jpg')
Out[28]:
No description has been provided for this image
In [30]:
# 모델 테스트
char_path = './test/kokomi.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 265ms/step
이 캐릭터는 [ Lumine ] 같습니다. 추정 확률은 91.63 % 입니다.
In [ ]:
# 데이터 추가 필요 ( 데이터 부족이 원인같음 )
In [32]:
import pathlib
data_dir = 'C:\\Users\\starl\\Downloads\\class\\data3\\'
data_dir = pathlib.Path(data_dir).with_suffix('')
In [33]:
image_count = len(list(data_dir.glob('*/*.jpg')))
print(image_count)
# 데이터 사본의 총 개수

400
In [36]:
import random 
import os
hu_tao = list(data_dir.glob('Hu Tao/*'))
PIL.Image.open(str(hu_tao[random.randint(1, len(os.listdir('./data3/Hu Tao')))]))
# 랜덤 이미지 추출
Out[36]:
No description has been provided for this image
In [38]:
import random 
import os
kokomi = list(data_dir.glob('kokomi/*'))
PIL.Image.open(str(kokomi[random.randint(1, len(os.listdir('./data3/kokomi')))]))
# 랜덤 이미지 추출
Out[38]:
No description has been provided for this image
In [39]:
# 데이터셋 생성 ( 매개변수 지정 )
batch_size = 32
img_height = 180
img_width = 180

train_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="training",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

Found 400 files belonging to 4 classes.
Using 320 files for training.
In [40]:
val_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="validation",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

Found 400 files belonging to 4 classes.
Using 80 files for validation.
In [41]:
class_names = train_ds.class_names
print(class_names)

['Albedo', 'Ayaka', 'Hu Tao', 'Kokomi']
In [46]:
import matplotlib.pyplot as plt

plt.figure(figsize=(10, 10))
for images, labels in train_ds.take(1):
  for i in range(9):
    ax = plt.subplot(5, 10, i + 1)
    plt.imshow(images[i].numpy().astype("uint8"))
    plt.title(class_names[labels[i]])
    plt.axis("off")
    # 훈련 데이터세트 출력
No description has been provided for this image
In [47]:
for image_batch, labels_batch in train_ds:
  print(image_batch.shape)
  print(labels_batch.shape)
  break
# image_batch는 (32, 180, 180, 3) 형상의 텐서

(32, 180, 180, 3)
(32,)
In [48]:
# 데이터세트 구성
AUTOTUNE = tf.data.AUTOTUNE

train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
In [49]:
# tf.keras.layers.Rescaling을 사용하여 값을 [0, 1] 범위로 표준화
normalization_layer = layers.Rescaling(1./255)
In [50]:
normalized_ds = train_ds.map(lambda x, y: (normalization_layer(x), y))
image_batch, labels_batch = next(iter(normalized_ds))
first_image = image_batch[0]
print(np.min(first_image), np.max(first_image))
# Dataset.map을 호출하여 데이터세트에 이를 적용

0.0 1.0
In [51]:
# 모델 생성
num_classes = len(class_names)

model = Sequential([
  layers.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(64, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Flatten(),
  layers.Dense(128, activation='relu'),
  layers.Dense(num_classes)
])
# relu 활성화 함수를 통해서 레이어 생성
In [52]:
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
# 아담 옵티마이저와 tf.keras.losses.SparseCategoricalCrossentropy 손실 함수를 선택
In [53]:
model.summary()
# 네트워크의 모든 레이어 호출( df['total_bill'].info() 같은 메서드 )

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 rescaling_3 (Rescaling)     (None, 180, 180, 3)       0         
                                                                 
 conv2d_3 (Conv2D)           (None, 180, 180, 16)      448       
                                                                 
 max_pooling2d_3 (MaxPoolin  (None, 90, 90, 16)        0         
 g2D)                                                            
                                                                 
 conv2d_4 (Conv2D)           (None, 90, 90, 32)        4640      
                                                                 
 max_pooling2d_4 (MaxPoolin  (None, 45, 45, 32)        0         
 g2D)                                                            
                                                                 
 conv2d_5 (Conv2D)           (None, 45, 45, 64)        18496     
                                                                 
 max_pooling2d_5 (MaxPoolin  (None, 22, 22, 64)        0         
 g2D)                                                            
                                                                 
 flatten_1 (Flatten)         (None, 30976)             0         
                                                                 
 dense_2 (Dense)             (None, 128)               3965056   
                                                                 
 dense_3 (Dense)             (None, 4)                 516       
                                                                 
=================================================================
Total params: 3989156 (15.22 MB)
Trainable params: 3989156 (15.22 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
In [54]:
# 100에포크 동안 모델 학습
epochs=100
history = model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs
)

Epoch 1/100
10/10 [==============================] - 10s 887ms/step - loss: 2.1312 - accuracy: 0.3531 - val_loss: 1.3565 - val_accuracy: 0.3625
Epoch 2/100
10/10 [==============================] - 8s 778ms/step - loss: 1.2047 - accuracy: 0.4219 - val_loss: 1.2171 - val_accuracy: 0.4500
Epoch 3/100
10/10 [==============================] - 8s 766ms/step - loss: 0.9867 - accuracy: 0.5281 - val_loss: 0.9715 - val_accuracy: 0.5125
Epoch 4/100
10/10 [==============================] - 8s 770ms/step - loss: 0.7383 - accuracy: 0.6875 - val_loss: 0.8079 - val_accuracy: 0.6500
Epoch 5/100
10/10 [==============================] - 9s 871ms/step - loss: 0.5383 - accuracy: 0.8000 - val_loss: 0.6835 - val_accuracy: 0.7375
Epoch 6/100
10/10 [==============================] - 8s 755ms/step - loss: 0.4184 - accuracy: 0.8594 - val_loss: 0.7421 - val_accuracy: 0.7250
Epoch 7/100
10/10 [==============================] - 10s 1s/step - loss: 0.2911 - accuracy: 0.9250 - val_loss: 0.8096 - val_accuracy: 0.6875
Epoch 8/100
10/10 [==============================] - 8s 811ms/step - loss: 0.2162 - accuracy: 0.9375 - val_loss: 0.7465 - val_accuracy: 0.7500
Epoch 9/100
10/10 [==============================] - 9s 958ms/step - loss: 0.1628 - accuracy: 0.9438 - val_loss: 0.8327 - val_accuracy: 0.7500
Epoch 10/100
10/10 [==============================] - 8s 844ms/step - loss: 0.1079 - accuracy: 0.9719 - val_loss: 0.8495 - val_accuracy: 0.7750
Epoch 11/100
10/10 [==============================] - 8s 840ms/step - loss: 0.0769 - accuracy: 0.9875 - val_loss: 0.5986 - val_accuracy: 0.7750
Epoch 12/100
10/10 [==============================] - 9s 906ms/step - loss: 0.0620 - accuracy: 0.9812 - val_loss: 0.6603 - val_accuracy: 0.7875
Epoch 13/100
10/10 [==============================] - 11s 1s/step - loss: 0.0609 - accuracy: 0.9781 - val_loss: 0.6658 - val_accuracy: 0.7750
Epoch 14/100
10/10 [==============================] - 8s 829ms/step - loss: 0.0870 - accuracy: 0.9719 - val_loss: 0.9394 - val_accuracy: 0.7250
Epoch 15/100
10/10 [==============================] - 7s 729ms/step - loss: 0.0934 - accuracy: 0.9812 - val_loss: 0.5432 - val_accuracy: 0.8000
Epoch 16/100
10/10 [==============================] - 12s 1s/step - loss: 0.0401 - accuracy: 0.9906 - val_loss: 0.6761 - val_accuracy: 0.7875
Epoch 17/100
10/10 [==============================] - 9s 873ms/step - loss: 0.0449 - accuracy: 0.9875 - val_loss: 0.7281 - val_accuracy: 0.7625
Epoch 18/100
10/10 [==============================] - 10s 976ms/step - loss: 0.0301 - accuracy: 0.9969 - val_loss: 0.7653 - val_accuracy: 0.7875
Epoch 19/100
10/10 [==============================] - 10s 1s/step - loss: 0.0123 - accuracy: 0.9969 - val_loss: 0.8545 - val_accuracy: 0.7750
Epoch 20/100
10/10 [==============================] - 9s 863ms/step - loss: 0.0054 - accuracy: 1.0000 - val_loss: 1.0139 - val_accuracy: 0.8125
Epoch 21/100
10/10 [==============================] - 10s 1s/step - loss: 0.0035 - accuracy: 1.0000 - val_loss: 0.8483 - val_accuracy: 0.8000
Epoch 22/100
10/10 [==============================] - 9s 886ms/step - loss: 0.0021 - accuracy: 1.0000 - val_loss: 0.9052 - val_accuracy: 0.8125
Epoch 23/100
10/10 [==============================] - 7s 683ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 0.9049 - val_accuracy: 0.8000
Epoch 24/100
10/10 [==============================] - 7s 678ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.9681 - val_accuracy: 0.8000
Epoch 25/100
10/10 [==============================] - 7s 683ms/step - loss: 0.0011 - accuracy: 1.0000 - val_loss: 0.9836 - val_accuracy: 0.8000
Epoch 26/100
10/10 [==============================] - 7s 683ms/step - loss: 9.5899e-04 - accuracy: 1.0000 - val_loss: 0.9885 - val_accuracy: 0.8000
Epoch 27/100
10/10 [==============================] - 7s 667ms/step - loss: 8.5743e-04 - accuracy: 1.0000 - val_loss: 1.0054 - val_accuracy: 0.8000
Epoch 28/100
10/10 [==============================] - 7s 670ms/step - loss: 7.7261e-04 - accuracy: 1.0000 - val_loss: 1.0174 - val_accuracy: 0.8125
Epoch 29/100
10/10 [==============================] - 7s 687ms/step - loss: 7.0321e-04 - accuracy: 1.0000 - val_loss: 1.0368 - val_accuracy: 0.8000
Epoch 30/100
10/10 [==============================] - 7s 680ms/step - loss: 6.4278e-04 - accuracy: 1.0000 - val_loss: 1.0408 - val_accuracy: 0.8125
Epoch 31/100
10/10 [==============================] - 7s 679ms/step - loss: 5.9647e-04 - accuracy: 1.0000 - val_loss: 1.0584 - val_accuracy: 0.8000
Epoch 32/100
10/10 [==============================] - 7s 670ms/step - loss: 5.5382e-04 - accuracy: 1.0000 - val_loss: 1.0733 - val_accuracy: 0.8125
Epoch 33/100
10/10 [==============================] - 7s 715ms/step - loss: 5.1674e-04 - accuracy: 1.0000 - val_loss: 1.0793 - val_accuracy: 0.8125
Epoch 34/100
10/10 [==============================] - 7s 685ms/step - loss: 4.9771e-04 - accuracy: 1.0000 - val_loss: 1.0786 - val_accuracy: 0.8125
Epoch 35/100
10/10 [==============================] - 8s 770ms/step - loss: 4.4487e-04 - accuracy: 1.0000 - val_loss: 1.0977 - val_accuracy: 0.8000
Epoch 36/100
10/10 [==============================] - 7s 705ms/step - loss: 4.1632e-04 - accuracy: 1.0000 - val_loss: 1.1119 - val_accuracy: 0.8125
Epoch 37/100
10/10 [==============================] - 7s 753ms/step - loss: 3.9250e-04 - accuracy: 1.0000 - val_loss: 1.1244 - val_accuracy: 0.8250
Epoch 38/100
10/10 [==============================] - 8s 773ms/step - loss: 3.6459e-04 - accuracy: 1.0000 - val_loss: 1.1309 - val_accuracy: 0.8000
Epoch 39/100
10/10 [==============================] - 7s 710ms/step - loss: 3.4460e-04 - accuracy: 1.0000 - val_loss: 1.1454 - val_accuracy: 0.8250
Epoch 40/100
10/10 [==============================] - 7s 735ms/step - loss: 3.2113e-04 - accuracy: 1.0000 - val_loss: 1.1484 - val_accuracy: 0.8000
Epoch 41/100
10/10 [==============================] - 7s 700ms/step - loss: 3.0817e-04 - accuracy: 1.0000 - val_loss: 1.1519 - val_accuracy: 0.8250
Epoch 42/100
10/10 [==============================] - 7s 684ms/step - loss: 2.9383e-04 - accuracy: 1.0000 - val_loss: 1.1778 - val_accuracy: 0.8000
Epoch 43/100
10/10 [==============================] - 7s 690ms/step - loss: 2.7234e-04 - accuracy: 1.0000 - val_loss: 1.1823 - val_accuracy: 0.8250
Epoch 44/100
10/10 [==============================] - 7s 682ms/step - loss: 2.6219e-04 - accuracy: 1.0000 - val_loss: 1.1951 - val_accuracy: 0.8000
Epoch 45/100
10/10 [==============================] - 7s 688ms/step - loss: 2.4925e-04 - accuracy: 1.0000 - val_loss: 1.1941 - val_accuracy: 0.8125
Epoch 46/100
10/10 [==============================] - 7s 679ms/step - loss: 2.3201e-04 - accuracy: 1.0000 - val_loss: 1.2163 - val_accuracy: 0.8000
Epoch 47/100
10/10 [==============================] - 7s 677ms/step - loss: 2.2107e-04 - accuracy: 1.0000 - val_loss: 1.2114 - val_accuracy: 0.8125
Epoch 48/100
10/10 [==============================] - 7s 681ms/step - loss: 2.1418e-04 - accuracy: 1.0000 - val_loss: 1.2288 - val_accuracy: 0.8000
Epoch 49/100
10/10 [==============================] - 7s 681ms/step - loss: 2.0060e-04 - accuracy: 1.0000 - val_loss: 1.2399 - val_accuracy: 0.8125
Epoch 50/100
10/10 [==============================] - 7s 683ms/step - loss: 1.9251e-04 - accuracy: 1.0000 - val_loss: 1.2572 - val_accuracy: 0.8125
Epoch 51/100
10/10 [==============================] - 7s 684ms/step - loss: 1.8769e-04 - accuracy: 1.0000 - val_loss: 1.2375 - val_accuracy: 0.8125
Epoch 52/100
10/10 [==============================] - 7s 673ms/step - loss: 1.8086e-04 - accuracy: 1.0000 - val_loss: 1.2541 - val_accuracy: 0.8000
Epoch 53/100
10/10 [==============================] - 7s 682ms/step - loss: 1.6912e-04 - accuracy: 1.0000 - val_loss: 1.2670 - val_accuracy: 0.8125
Epoch 54/100
10/10 [==============================] - 7s 679ms/step - loss: 1.6195e-04 - accuracy: 1.0000 - val_loss: 1.2861 - val_accuracy: 0.8125
Epoch 55/100
10/10 [==============================] - 7s 673ms/step - loss: 1.5139e-04 - accuracy: 1.0000 - val_loss: 1.2966 - val_accuracy: 0.8125
Epoch 56/100
10/10 [==============================] - 7s 664ms/step - loss: 1.4818e-04 - accuracy: 1.0000 - val_loss: 1.2996 - val_accuracy: 0.8125
Epoch 57/100
10/10 [==============================] - 7s 666ms/step - loss: 1.4365e-04 - accuracy: 1.0000 - val_loss: 1.2896 - val_accuracy: 0.8125
Epoch 58/100
10/10 [==============================] - 7s 666ms/step - loss: 1.4148e-04 - accuracy: 1.0000 - val_loss: 1.3195 - val_accuracy: 0.8000
Epoch 59/100
10/10 [==============================] - 7s 697ms/step - loss: 1.3122e-04 - accuracy: 1.0000 - val_loss: 1.3092 - val_accuracy: 0.8125
Epoch 60/100
10/10 [==============================] - 7s 668ms/step - loss: 1.2192e-04 - accuracy: 1.0000 - val_loss: 1.3234 - val_accuracy: 0.8125
Epoch 61/100
10/10 [==============================] - 7s 667ms/step - loss: 1.1763e-04 - accuracy: 1.0000 - val_loss: 1.3320 - val_accuracy: 0.8125
Epoch 62/100
10/10 [==============================] - 7s 686ms/step - loss: 1.1307e-04 - accuracy: 1.0000 - val_loss: 1.3431 - val_accuracy: 0.8125
Epoch 63/100
10/10 [==============================] - 7s 665ms/step - loss: 1.1102e-04 - accuracy: 1.0000 - val_loss: 1.3443 - val_accuracy: 0.8125
Epoch 64/100
10/10 [==============================] - 7s 665ms/step - loss: 1.0566e-04 - accuracy: 1.0000 - val_loss: 1.3433 - val_accuracy: 0.8125
Epoch 65/100
10/10 [==============================] - 7s 664ms/step - loss: 1.0153e-04 - accuracy: 1.0000 - val_loss: 1.3594 - val_accuracy: 0.8125
Epoch 66/100
10/10 [==============================] - 7s 679ms/step - loss: 9.8235e-05 - accuracy: 1.0000 - val_loss: 1.3534 - val_accuracy: 0.8125
Epoch 67/100
10/10 [==============================] - 7s 668ms/step - loss: 9.3993e-05 - accuracy: 1.0000 - val_loss: 1.3632 - val_accuracy: 0.8125
Epoch 68/100
10/10 [==============================] - 7s 667ms/step - loss: 9.2462e-05 - accuracy: 1.0000 - val_loss: 1.3800 - val_accuracy: 0.8125
Epoch 69/100
10/10 [==============================] - 7s 671ms/step - loss: 8.9840e-05 - accuracy: 1.0000 - val_loss: 1.3784 - val_accuracy: 0.7875
Epoch 70/100
10/10 [==============================] - 7s 670ms/step - loss: 8.6830e-05 - accuracy: 1.0000 - val_loss: 1.3884 - val_accuracy: 0.8125
Epoch 71/100
10/10 [==============================] - 7s 664ms/step - loss: 8.2839e-05 - accuracy: 1.0000 - val_loss: 1.3868 - val_accuracy: 0.8125
Epoch 72/100
10/10 [==============================] - 7s 665ms/step - loss: 7.9219e-05 - accuracy: 1.0000 - val_loss: 1.3919 - val_accuracy: 0.8125
Epoch 73/100
10/10 [==============================] - 7s 676ms/step - loss: 7.6702e-05 - accuracy: 1.0000 - val_loss: 1.3965 - val_accuracy: 0.8125
Epoch 74/100
10/10 [==============================] - 7s 669ms/step - loss: 7.4049e-05 - accuracy: 1.0000 - val_loss: 1.4053 - val_accuracy: 0.8125
Epoch 75/100
10/10 [==============================] - 7s 669ms/step - loss: 7.2392e-05 - accuracy: 1.0000 - val_loss: 1.4039 - val_accuracy: 0.8125
Epoch 76/100
10/10 [==============================] - 7s 668ms/step - loss: 7.1185e-05 - accuracy: 1.0000 - val_loss: 1.4180 - val_accuracy: 0.8125
Epoch 77/100
10/10 [==============================] - 7s 674ms/step - loss: 7.1389e-05 - accuracy: 1.0000 - val_loss: 1.4133 - val_accuracy: 0.7875
Epoch 78/100
10/10 [==============================] - 7s 662ms/step - loss: 6.8864e-05 - accuracy: 1.0000 - val_loss: 1.4345 - val_accuracy: 0.8125
Epoch 79/100
10/10 [==============================] - 7s 676ms/step - loss: 6.4980e-05 - accuracy: 1.0000 - val_loss: 1.4264 - val_accuracy: 0.8125
Epoch 80/100
10/10 [==============================] - 7s 670ms/step - loss: 6.3597e-05 - accuracy: 1.0000 - val_loss: 1.4327 - val_accuracy: 0.8125
Epoch 81/100
10/10 [==============================] - 7s 670ms/step - loss: 6.0305e-05 - accuracy: 1.0000 - val_loss: 1.4371 - val_accuracy: 0.8000
Epoch 82/100
10/10 [==============================] - 7s 674ms/step - loss: 5.8236e-05 - accuracy: 1.0000 - val_loss: 1.4549 - val_accuracy: 0.8125
Epoch 83/100
10/10 [==============================] - 7s 665ms/step - loss: 5.6330e-05 - accuracy: 1.0000 - val_loss: 1.4480 - val_accuracy: 0.8000
Epoch 84/100
10/10 [==============================] - 7s 668ms/step - loss: 5.4761e-05 - accuracy: 1.0000 - val_loss: 1.4597 - val_accuracy: 0.8125
Epoch 85/100
10/10 [==============================] - 7s 667ms/step - loss: 5.3698e-05 - accuracy: 1.0000 - val_loss: 1.4605 - val_accuracy: 0.8125
Epoch 86/100
10/10 [==============================] - 7s 685ms/step - loss: 5.3100e-05 - accuracy: 1.0000 - val_loss: 1.4673 - val_accuracy: 0.8000
Epoch 87/100
10/10 [==============================] - 7s 671ms/step - loss: 5.2338e-05 - accuracy: 1.0000 - val_loss: 1.4642 - val_accuracy: 0.8000
Epoch 88/100
10/10 [==============================] - 7s 668ms/step - loss: 5.2763e-05 - accuracy: 1.0000 - val_loss: 1.4803 - val_accuracy: 0.8125
Epoch 89/100
10/10 [==============================] - 7s 664ms/step - loss: 5.2411e-05 - accuracy: 1.0000 - val_loss: 1.4695 - val_accuracy: 0.7875
Epoch 90/100
10/10 [==============================] - 7s 670ms/step - loss: 5.1189e-05 - accuracy: 1.0000 - val_loss: 1.4891 - val_accuracy: 0.8125
Epoch 91/100
10/10 [==============================] - 7s 666ms/step - loss: 4.6897e-05 - accuracy: 1.0000 - val_loss: 1.4881 - val_accuracy: 0.8125
Epoch 92/100
10/10 [==============================] - 7s 674ms/step - loss: 4.6245e-05 - accuracy: 1.0000 - val_loss: 1.4922 - val_accuracy: 0.7875
Epoch 93/100
10/10 [==============================] - 7s 665ms/step - loss: 4.4929e-05 - accuracy: 1.0000 - val_loss: 1.5004 - val_accuracy: 0.8125
Epoch 94/100
10/10 [==============================] - 7s 666ms/step - loss: 4.2745e-05 - accuracy: 1.0000 - val_loss: 1.5092 - val_accuracy: 0.8000
Epoch 95/100
10/10 [==============================] - 7s 672ms/step - loss: 4.1048e-05 - accuracy: 1.0000 - val_loss: 1.5077 - val_accuracy: 0.8000
Epoch 96/100
10/10 [==============================] - 7s 671ms/step - loss: 3.9749e-05 - accuracy: 1.0000 - val_loss: 1.5135 - val_accuracy: 0.8000
Epoch 97/100
10/10 [==============================] - 7s 665ms/step - loss: 3.8849e-05 - accuracy: 1.0000 - val_loss: 1.5202 - val_accuracy: 0.8000
Epoch 98/100
10/10 [==============================] - 7s 666ms/step - loss: 3.7774e-05 - accuracy: 1.0000 - val_loss: 1.5234 - val_accuracy: 0.8000
Epoch 99/100
10/10 [==============================] - 7s 681ms/step - loss: 3.6858e-05 - accuracy: 1.0000 - val_loss: 1.5293 - val_accuracy: 0.8000
Epoch 100/100
10/10 [==============================] - 7s 673ms/step - loss: 3.6018e-05 - accuracy: 1.0000 - val_loss: 1.5304 - val_accuracy: 0.8000
In [55]:
model.save('./models/genshin2.HDF5')

INFO:tensorflow:Assets written to: ./models/genshin2.HDF5\assets

INFO:tensorflow:Assets written to: ./models/genshin2.HDF5\assets
In [56]:
# 훈련 결과 시각화
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']

loss = history.history['loss']
val_loss = history.history['val_loss']

epochs_range = range(epochs)

plt.figure(figsize=(8, 8))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
No description has been provided for this image
In [57]:
PIL.Image.open('./test/kokomi.jpg')
Out[57]:
No description has been provided for this image
In [58]:
# 모델 테스트
char_path = './test/kokomi.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 130ms/step
이 캐릭터는 [ Kokomi ] 같습니다. 추정 확률은 100.00 % 입니다.
In [59]:
PIL.Image.open('./test/albedo.jpg')
Out[59]:
No description has been provided for this image
In [60]:
# 모델 테스트
char_path = './test/albedo.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 34ms/step
이 캐릭터는 [ Albedo ] 같습니다. 추정 확률은 99.64 % 입니다.
In [61]:
PIL.Image.open('./test/ayaka.jpg')
Out[61]:
No description has been provided for this image
In [62]:
# 모델 테스트
char_path = './test/ayaka.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 72ms/step
이 캐릭터는 [ Ayaka ] 같습니다. 추정 확률은 100.00 % 입니다.
In [67]:
PIL.Image.open('./test/hutao2.jpg')
Out[67]:
No description has been provided for this image
In [68]:
# 모델 테스트
char_path = './test/hutao2.jpg'

img = tf.keras.utils.load_img(
    char_path, target_size=(img_height, img_width)
)
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # 배치 생성

predictions = model.predict(img_array)
score = tf.nn.softmax(predictions[0])

print(
    "이 캐릭터는 [ {} ] 같습니다. 추정 확률은 {:.2f} % 입니다."
    .format(class_names[np.argmax(score)], 100 * np.max(score))
)

1/1 [==============================] - 0s 41ms/step
이 캐릭터는 [ Hu Tao ] 같습니다. 추정 확률은 100.00 % 입니다.